A model that answers is not a system that runs.
The distance between them is permission.review.ownership.accountability.
Tools for art teams ⟶ tools for developers ⟶ systems people actually run.
Acting CTO of an AI platform · 5 industry internships · 2 startups · Studied in 3 countries · 2 ACM papers
Hobart, Tasmaniajacobharrisau · CC BY-SA 2.0About Me
How I got here, and what I am useful for.I build and test software other people have to rely on: IT, data and AI systems.
5 internships · 4 volunteer roles · 3 countries · 2 startups · 1 of me
I'm finishing a Master of Information Technology and Systems at the University of Tasmania. My experience covers AI agent engineering, data and business analytics, interactive and simulation technology, and web development: four paid full-time internships, two startups I co-founded, and an IT internship in Australia now.
At an AI developer-tools company I did AI agent development, integrating and testing AI coding assistants in a Cursor-style IDE — the hard part was the prompt, the context and the tests, not the model. At Tencent, one of the world's largest technology and gaming companies, I analysed gameplay data and prototyped game logic for Honor of Kings; before that I built tools and content pipelines for art teams at Perfect World and Lilith Games.
I co-founded an augmented-reality company planning location-based AR games for shopping centres and tourist sites, and led the AI content side up to a seed-round plan. Later I co-founded a live-streaming media startup, where I ran recruitment and onboarding for a team of 20 — screening 300+ people taught me to write a requirement people can actually answer.
I volunteer the same way I work. At Glenorchy Library I give one-to-one digital help — the cheapest usability testing there is. At children's STEAMM workshops in Rokeby I test the robots and iPads first, then hand the children their own Blockly errors to debug. I'm also on the Hospitals South consumer register. Two more start in 2027: youth mentoring at Hobart High School through Raise, and local sponsorship for the Mother's Day Classic.
Right now I'm acting CTO at AIA-Institute, building an AI platform for Australian organisations; it is pre-release, so there is nothing more about it here. Alongside it I run the IT systems work at Strategy&Ops Consultancy and do LiDAR perception with UTAS Motorsport. Everything on this page comes from one belief: a model that can answer a question is not yet a system an organisation can run.
I'm open to graduate and junior roles, part-time work, placements and internships in Tasmania, in software development, data and business analysis, or interactive and simulation technology. If any of this is useful to you, I'd be glad to talk.
What I'm Good At
Pick an area. The panel shows where it was actually used.I wrote the code, supported it inside other people's organisations, and now spend most of my time where the systems meet the people who have to run them.
Numbers somebody has to act on.
- Tencent
- Win rates, pick rates, skill effectiveness and backend telemetry, read closely enough to inform the parameters a live game was balanced on.
- AgriSentry
- Simulated weather and sensor data turned into per-crop risk alerts on the Power Platform, with a Power BI dashboard over an immutable audit trail.
- Malou Media
- 300+ candidates screened, with regular reporting on candidate sources, conversion rates and time-to-hire.
What I'm Building
Drag them around. Tap one to open it.Rivulet LocalOps

Turns a written service policy into decisions that can be executed, checked and traced to a named owner. The reference scenario is a council handling waste and litter requests.
AI-assisted and policy-gated: a model reads the request and suggests type and urgency, but the written policy decides what may happen — it may raise urgency and never lower it — and a named person makes the final call.
Every request is diagnosed, tested against the approved policy, recorded in a tamper-evident ledger and reviewed by a second person.
An independent prototype, not an official council service: no real resident requests, and no writes to any real council system.
LifeOps Atlas

A single-user operations system that brings tasks, relationships, event research and career preparation into one evidence-backed workspace.
Durable records live in PostgreSQL with append-only corrections and source evidence; the AI works only from bounded, traceable input.
Node.js 24, TypeScript, Fastify 5, Zod, PostgreSQL 17 and Docker Compose. The public edition is a sanitised walkthrough built from synthetic records.
AgriSentry

A configurable farm-monitoring system on the Microsoft Power Platform: simulated weather and sensor data become per-crop risk alerts.
A person approves every alert before anything happens, with an immutable audit trail behind it and a Power BI dashboard over it.
Power Apps, Dataverse, Power Automate and Power BI with an Azure Functions rule engine. Synthetic data; the application source stays private.
Evan.H Portfolio

This site. A portfolio has two readers — someone skimming for facts, and someone who wants to look around — so it is built twice over one set of content.
Every word lives in content/*.json behind a typed loader that fails the build on a broken link, an invented status or a second language, with 104 tests behind it.
The same content renders as this page and as an explorable canvas deck with a voice guide that can only say what the slides say.
Tasmania Student Notes

An independent, student-led publication for Tasmania in six languages, with a newsroom of four AI roles where a human editor approves every final result.
A six-language preview site on Ghost's Headline theme, with a pixel office that replays recorded pilot runs.
The newsroom — Coordinator, Researcher, Writer and Checker — runs privately in Docker on the editor's own computer; the Coordinator handles routine hand-offs and the editor approves every published result.
A prototype with no published stories yet: the pilot runs used fictional material.
AI-Native Payment & Agent Engineering Lab

A payment-processing prototype and a read-only operations copilot, built to find where agent workflows break rather than to demo where they work.
Designed a FastAPI/SQLite payment-processing prototype normalising synchronous, webhook and polling results from three simulated channels through adapters, event de-duplication, audit logging and a deterministic state machine.
Verified a read-only RAG operations copilot with three tools and a Planner–Developer–Tester–Reviewer process, using 77 automated tests, 11 documented failure cases, human approval and code-level safety gates.
Cyber Finance City

Modular UE5 financial-education system with AI-driven NPCs and a FastAPI + RAG backend.
Created a modular UE5 interactive experience with AI-driven NPCs, configurable questions, scoring workflows, PBR environments, and VR-ready interactions.
Connected the UE5 client to FastAPI, structured question data, RAG retrieval, and LLM services, then validated API, integration, and end-to-end interaction workflows.
Little Red: Sequel

Independently developed story-driven 3D pixel-style Unreal Engine indie game demo.
Developed a story-driven 3D pixel-style Unreal Engine game demo featuring dark-fantasy worldbuilding, hub-based exploration, nonlinear quests, interactive dialogue, and multiple endings.
Modified and combined Marketplace quest, dialogue and menu systems with custom Blueprints, and built the AIGC-assisted asset production that put 2D pixel art onto 3D environments.
Case Studies
Six narrated walkthroughs. Synthetic data throughout.
Rivulet LocalOps
One synthetic request through ten written checks: where it is, whose jurisdiction it falls in, what the policy allows, and who signed it off.

LifeOps Atlas
What the system checked and what it refused to guess, ending on the numbers from the side-by-side trial rather than on a claim.

AgriSentry
A farm AI that watches, explains in plain words, and still waits for a person before anything is sprayed.

AI-Native Payment & Agent Engineering Lab
An investigation into a payment agent that passed everything it was asked to pass, and what that still did not prove.

Cyber Finance City
A walk through the UE5 city, where the questions the characters ask come from a retrieval service rather than from the level.

Little Red: Sequel
A solo Unreal Engine game whose pixel-style art came off a Stable Diffusion pipeline rather than being drawn one asset at a time.
Where I'm Heading
The page above is what exists. This is what I'm aiming at.Written as intent, not as experience — nothing here is a claim about work already done. The detail is still being filled in.
What I want to do
The roles this work is pointed at.
What I want to be good at
The ground I want to be able to stand on.
My Values
How I actually work with people.More autonomy, more accountability
The more autonomy an agent has, the more identity, permissions, data boundaries, approvals, monitoring, traceability and human accountability matter.
Human judgement sets the value
AI should extend professional knowledge, not replace creative judgement: human taste, brand, mechanics and distinctive input decide what the output is worth.
Say what is tested, and what is assumed
A prototype should say that it is a prototype. I keep what has been tested apart from what is still assumed: failure cases get written down, limits go on the page rather than in a footnote, and a pilot that has published nothing says so.
My Record
Where I have worked, and where I studied.Work
Open a row for what the work actually was.AIA-InstituteEarly-stage AI platformChief Technology Officer (Acting)Technical lead on a pre-release AI platform
- Leading the technical design and build of an AI platform for Australian organisations, working with the CEO and co-founder.
- The platform is pre-release, so there is no detail about it here.
Strategy&Ops ConsultancyStrategy and operations consultancyIT Systems Intern18-member directory shipped · 300+ pages and 84 lessons audited
- Redesigned and launched the company's Net-Zero Global Network page in WordPress: a searchable, filterable directory of 18 member organisations with an auto-rotating carousel, built from a tested prototype.
- Maintained the WordPress site and Tutor LMS under change control — staging tests, approval steps, a shared change log, live checks on desktop and mobile — and rebuilt the homepage hero slider.
- Audited the videos embedded across 300+ public pages and 84 lessons in 20 LMS courses, and fixed playback times on 55 video lessons.
Qingqiu Technology Co., Ltd.AI developer-tooling startupSoftware Engineer InternA Cursor-style AI coding IDE · Codex and Claude in the developer workflow
- Developed a Cursor-style AI coding IDE for an indie game studio.
- Integrated OpenAI Codex and Anthropic Claude into developer workflows, writing the system prompts, few-shot examples, and the context-management and model-interaction logic.
- Built context-processing and web workflow features in JavaScript/Node.js, and tested multi-turn conversations, prompt behaviour, process logic and the generated code.
TencentWorld's largest games companyTechnical Game Systems InternTelemetry read to support balancing on Honor of Kings
- Analysed win rates, pick rates, skill effectiveness, player feedback and backend telemetry to support gameplay balancing and parameter adjustments in Honor of Kings.
- Modelled and iterated skill configurations and numerical parameters in Excel and internal simulation tools, including damage, healing, cooldowns, mana costs, movement decay, and effect stacking.
- Prototyped and debugged Lua gameplay modules and collaborated with engineers on state-machine restructuring, interaction validation, multiplayer testing, and VFX/SFX/animation synchronisation.
Malou Media Ltd (Startup)Startup · 20 peopleCo-Founder & Recruitment Manager300+ candidates screened · recruitment run single-handed for 20 people
- Independently ran recruitment, onboarding and people operations for a 20-person company spanning content planning, livestreaming and operations.
- Screened 300+ candidates and standardised the end-to-end recruitment process, using regular reporting to track candidate sources, conversion rates, and time-to-hire.
Perfect World Co., Ltd.Listed games publisherAI Workflow Engineering InternLoRA fine-tuning · 2D/3D dataset curation for an internationally released title
- Built and maintained 2D/3D asset datasets for an internationally released game, collecting, annotating and style-tagging characters and props.
- Trained and refined LoRA models with Stable Diffusion and ComfyUI, and collaborated on Python/Java/C++ automation via Jenkins, Unity asset integration, technical documentation, and troubleshooting.
AR Platform Ltd (Startup)Startup · co-foundedCo-Founder & AIGC Technical Lead4–5 person team · 20+ external stakeholders · a seed-round business plan
- Co-founded a 4–5 person AR content startup as AIGC lead and designed the AI content pipeline, from 2D concept art to 3D assets, and the interaction-flow and spatial-mapping concepts for the location-based multiplayer AR games in its seed-round business plan.
- Coordinated with 20+ external stakeholders and prepared B2B use cases and commercial proposals for venue operators, including annual licensing, event marketing, and customer-engagement scenarios.
Lilith GamesAFK Journey · Rise of KingdomsAI Workflow Engineering InternAIGC pipelines across global live-service titles · training for non-technical art teams
- Automated Unity batch operations with Python and C#, streamlining asset import, classification, compression, and tagging while reducing repetitive manual processing.
- Built AIGC pipelines for AFK Journey and Rise of Kingdoms: dataset preprocessing, LoRA fine-tuning, and Stable Diffusion/ComfyUI integration with five DCC tools (Blender, Maya, Houdini, 3ds Max, Cinema 4D); documented them and trained non-technical art teams.
Education
A vocational diploma in China to a master's in Hobart — four institutions, two countries, two languages of instruction.Published
Co-authored, and the note says which part was mine.Tell me what keeps breaking.
Roles, projects, volunteering, or a coffee in Hobart. The messy ones are the interesting ones.
Away From the Desk
The part that does not fit on a resume.- Most weeks I sit with someone at Glenorchy Library who is trying to make a government website work on a phone they were handed last week. That is where you find out what a system is really like.
- I help run STEAMM robotics workshops for six to twelve year olds in Rokeby — mostly by testing every robot and tablet before the children arrive, which is its own kind of engineering.
- Nine ICT and professional events in one semester, with notes after each. A question I asked after SecTalks Hobart is how I ended up at an AISA Tasmania branch event as an invited guest.
- From 2027 I mentor a high school student through Raise. The training was about what a young person actually needs from a mentor, which turns out not to be advice.
- I tutor a classmate from a non-IT background across two units. Explaining a concept to someone who has no reason to already understand it is the fastest way to find out whether I do.






